Anthropic · 2026-08-14 · notable
Claude Code token costs — Anthropic explains what makes sessions expensive
Anthropic's guide explains what drives Claude Code token costs: stale context left in the conversation, mid-session model switches that break the prompt cache, and noisy command output. Cache reads cost 0.1x the input price.
Anthropic's guide lists the habits that make a Claude Code session cost more than it needs to.
What is it?
A guide by Lydia Hallie, published by Anthropic on August 14, 2026, maps where Claude Code tokens actually go. It works through concrete habits — clearing between tasks, compacting before a break, @-mentioning files instead of typing paths, and quieting commands that dump large output — rather than general advice about writing better prompts.
How does it work?
Every turn re-sends the whole conversation, so anything left in context gets paid for again and again. Prompt caching softens that: cache reads cost 0.1x the input price, but the cache expires after one hour on a subscription and five minutes on an API key. Changing /model or /effort mid-conversation invalidates the cache, which is why the guide says to set both at the start of a session. Running a noisy command inside a subagent keeps its output out of the main conversation.
Why does it matter?
One long Claude Code session costs more than several short ones, because each turn re-reads everything before it. Developers on metered API billing, and teams watching per-seat limits, get a checklist they can apply today: /clear between unrelated tasks, /compact before stepping away, and MAX_THINKING_TOKENS or BASH_MAX_OUTPUT_LENGTH when thinking and command output are the thing filling the window.
Who is it for?
Claude Code users on API billing or tight plan limits
Try it
/context — see what is filling your Claude Code context window